Automated classification of cytogenetic abnormalities in hematolymphoid neoplasms

Abstract

Motivation Algorithms for classifying chromosomes, like convolutional deep neural networks (CNNs), show promise to augment cytogeneticists’ workflows, however, a critical limitation is their inability to accurately classify various structural chromosomal abnormalities. In hematopathology, recurrent structural cytogenetic abnormalities herald diagnostic, prognostic, and therapeutic implications, but are laborious for expert cytogeneticists to identify. Non-recurrent cytogenetic abnormalities also occur frequently cancerous cells. Here, we demonstrate the feasibility of using CNNs to accurately classify many recurrent cytogenetic abnormalities while being able to reliably detect non-recurrent, spurious abnormal chromosomes, as well as provide insights into dataset assembly, model selection, and training methodology that improve overall generalizability and performance for chromosome classification. Results Our top-performing model achieved a mean weighted F1 score of 96.86% on the validation set and 94.03% on the test set. Gradient class activation maps indicated that our model learned biologically-meaningful feature maps, reinforcing the clinical utility of our proposed approach. Altogether, this work: proposes a new dataset framework for training chromosome classifiers for use in a clinical environment, reveals that residual CNNs and cyclical learning rates confer superior performance, and demonstrates the feasibility of using this approach to automatically screen for many recurrent cytogenetic abnormalities while adeptly classifying non-recurrent abnormal chromosomes. Availability and Implementation Software is freely available at https://github.com/DaehwanKimLab/Chromosome-ReAd. The data underlying this article cannot be shared publicly due to it being protected patient information. Supplementary information Supplementary data are available at Bioinformatics online.

Optical Genome Mapping for Cytogenetic Diagnostics in AML

“…Considering the immense number of cells and potentially different malignant clones, this is a comparatively small number, which may not be representative of the entire sample. With the help of artificial intelligence (AI)-based programs, the time of analysis can be decreased and the number of analyzed metaphases increased [ 8 , 9 ]. Taken together, the method is nevertheless time consuming since a complete workup might require between 7 to 10 days under real-life conditions.…”

Applications of artificial intelligence in clinical laboratory genomics

“…Because AI methods have been developed for visual pattern recognition in X‐rays, computed tomography scans, and stained tissue slices, one might predict that these methods could also be applied to the analysis of chromosomal karyotypes for constitutional rearrangements or to the analysis of tumor tissue for chromosomal rearrangements. To date, little to no AI appears to be used to routinely analyze chromosomal karyotypes for constitutional rearrangements (Tseng et al, 2023), although various efforts have been used to decipher chromosomal rearrangements in cancer specimens, such as from karyotyped hematologic malignancies (Bokhari et al, 2022; Cox et al, 2022; Vajen et al, 2022; Walter et al, 2021). As genomic analysis increasingly shifts toward molecular approaches, even for chromosomal disorders, AI methods are being developed to identify chromosomal deletions, duplications, and other types of rearrangements from NGS data directly (Lin et al, 2022; Popic et al, 2023).…”

Automatic detection of a structural chromosomal abnormality: the del(5q) deletion

“…Computer vision techniques have been applied to various chromosome-related tasks, such as chromosome segmentation, 2 chromosome classification, 3 chromosome generation, 4 trisomy detection, 5 translocation detection, 6 and the detection of structural abnormalities. 7 These studies have shown promising results and should contribute to improving the analysis of chromosomal images.…”